PACE-RAG: Patient-Aware Contextual and Evidence-based Policy RAG for Clinical Drug Recommendation

Abstract

Drug recommendation requires a deep understanding of individual patient context, especially for complex conditions like Parkinson's disease. While LLMs possess broad medical knowledge, they fail to capture the subtle nuances of actual prescribing patterns. Existing RAG methods also struggle with these complexities because guideline-based retrieval remains too generic and similar-patient retrieval often replicates majority patterns without accounting for the unique clinical nuances of individual patients. To bridge this gap, we propose PACE-RAG (Patient-Aware Contextual and Evidence-based Policy RAG), a novel framework designed to synthesize individual patient context with the prescribing tendencies of similar cases. By analyzing treatment patterns tailored to specific clinical signals, PACE-RAG identifies optimal prescriptions and generates an explainable clinical summary. Evaluated on a Parkinson's cohort and the MIMIC-IV benchmark using Llama-3.1-8B and Qwen3-8B, PACE-RAG achieved state-of-the-art performance, reaching F1 scores of 80.84% and 47.22%, respectively. These results validate PACE-RAG as a robust, clinically grounded solution for personalized decision support. Our code is available at: https://github.com/ChaeYoungHuh/PACE-RAG.

Figures (15)

• Figure 1: Comparison of medical RAG approaches. While Guideline RAG leverages external medical knowledge, its recommendations tend to be generalized for specific clinical contexts. Although TreatRAG utilizes evidence-based medicine by referencing analogous patient data, it lacks individualization. PACE-RAG achieves personalized medicine by integrating a verification process with patient-similarity retrieval.
• Figure 2: Overview of the PACE-RAG framework architecture. The pipeline consists of four stages: (1) Focus-Specific Retrieval to identify similar patients from $x_t$; (2) Prescribing Tendency Analysis to extract medication patterns from retrieved cases; (3) Prescribing Refinement to verify the initial prescription against identified tendencies; and (4) Clinical Summary to synthesize the final treatment rationale.
• Figure 3: Specialist evaluation results. Comparative assessment of PACE-RAG and Guideline RAG by clinical specialists across six evaluation criteria.
• Figure 4: Relevance scores. The distribution of relevance scores between the extracted keywords and the ground truth (Mean=$3.79$, SD=$1.19$).
• Figure 5: Comparison of retrieval quality via drug precision@$k$. While TreatRAG accumulates irrelevant drugs as $k$ increases, PACE-RAG isolates symptom-specific interventions to maintain higher precision.